Deepfake pornography
Deepfake pornography, or simply fake pornography, is a type of synthetic porn that is created via altering already-existing pornographic material by applying deepfake technology to the faces of the actors. The use of deepfake porn has sparked controversy because it involves the making and sharing of realistic videos featuring non-consenting individuals, typically female celebrities, and is sometimes used for revenge porn. Efforts are being made to combat these ethical concerns through legislation and technology-based solutions.
Ethical considerations[edit]
Deepfake CSAM[edit]
Deepfake technology has made the creation of child sexual abuse material (CSAM), also often referenced to as child pornography, faster, safer and easier than it has ever been. Deepfakes can be used to produce new CSAM from already existing material or creating CSAM from children who have not been subjected to sexual abuse. Deepfake CSAM can, however, have real and direct implications on children including defamation, grooming, extortion, and bullying.[31]
Consent[edit]
Most of deepfake porn is made with faces of people who did not consent to their image being used in such a sexual way. In 2023, Sensity, an identify verification company, has found that "96% of deepfakes are sexually explicit and feature women who didn’t consent to the creation of the content."[32] Oftentimes, deepfake porn is used to humiliate and harass primarily women in ways similar to revenge porn.
Combatting deepfake pornography[edit]
Technical approach[edit]
Deepfake detection has become an increasingly important area of research in recent years as the spread of fake videos and images has become more prevalent. One promising approach to detecting deepfakes is through the use of Convolutional Neural Networks (CNNs), which have shown high accuracy in distinguishing between real and fake images. One CNN-based algorithm that has been developed specifically for deepfake detection is DeepRhythm, which has demonstrated an impressive accuracy score of 0.98 (i.e. successful at detecting deepfake images 98% of the time). This algorithm utilizes a pre-trained CNN to extract features from facial regions of interest and then applies a novel attention mechanism to identify discrepancies between the original and manipulated images. While the development of more sophisticated deepfake technology presents ongoing challenges to detection efforts, the high accuracy of algorithms like DeepRhythm offers a promising tool for identifying and mitigating the spread of harmful deepfakes.[33]
Aside from detection models, there are also video authenticating tools available to the public. In 2019, Deepware launched the first publicly available detection tool which allowed users to easily scan and detect deepfake videos. Similarly, in 2020 Microsoft released a free and user-friendly video authenticator. Users upload a suspected video or input a link, and receive a confidence score to assess the level of manipulation in a deepfake.
Legal approach[edit]
As of 2023, there is a lack of legislation that specifically addresses deepfake pornography. Instead, the harm caused by its creation and distribution is being addressed by the courts through existing criminal and civil laws.[1]
Victims of deepfake pornography often have claims for revenge porn, tort claims, and harassment.[34] The legal consequences for revenge porn vary from state to state and country to country.[35][36] For instance, in Canada, the penalty for publishing non-consensual intimate images is up to 5 years in prison,[37] whereas in Malta it is a fine of up to €5,000.[38]
The "Deepfake Accountability Act" was introduced to the United States Congress in 2019 but has died in 2020.[39] It aimed to make the production and distribution of digitally altered visual media that was not disclosed to be such, a criminal offense. The title specifies that making any sexual, non-consensual altered media with the intent of humiliating or otherwise harming the participants, may be fined, imprisoned for up to 5 years or both.[36] A newer version of bill was introduced in 2021 which would have required any "advanced technological false personation records" to contain a watermark and an audiovisual disclosure to identify and explain any altered audio and visual elements. The bill also includes that failure to disclose this information with intent to harass or humilitate a person with an "advanced technological false personation record" containing sexual content "shall be fined under this title, imprisoned for not more than 5 years, or both." However this bill has since died in 2023.[40]